IPv6 Transition & Operational Realitycenic08.cenic.org/program/slides/080312.cenic-v6-op-reality.pdfIPv6 Transition & Operational Reality CENIC 08 Oakland / 2008.03.12 Randy Bush

IPv6 Transition &Operational Reality

CENIC 08Oakland / 2008.03.12

Randy Bush <[email protected]>

http://rip.psg.com/~randy/080312.cenic-op-reality.pdf

2008.03.12 CENIC IPv6 Ops Reality Copyright 2007-8 RGnet, LLC 2

The Bottom Line

I am far less interested in what particular service is available via IPv4, IPv6, or both than I am in ensuring that it makes no difference to the users.


Just Deploy IPv6! If we were designing it again

we might do it differently But it is too late given IPv4

Free Pool run-out We have no choice, deploy

IPv6 or break the internet


Content• Some Hard Realities• Too Many IPv6 Myths• Strategic Problems in Transition• Tactical Transition Needs

There is no accompanying paper, so the slides are dense and meant as reference


Reality Therapy• The choice is massive IPv4 NAT or dual

stack with NAT-PT, get over it• IPv6 is a technically better more scalable

choice, get over it• The issues are when and how• Marketing fantasy is not helping us actually

deploy• Take off the ‘rose colored glasses’ so we

can see what reality is so we can actually make deployment decisions


What Should Have Happened

IPv6Deployment

IPv4Free Pool

$/IPv4/24

Today


What Is Happening?

IPv6Deployment

IPv4Free Pool

$/IPv4/24

Today

Optimism


Why Is This Happening?No transition planDeclared victory before hard part startedNo real long term planNo realistic estimation of costsNo support for the folk on the front linesVictory will be next monthThis Describes: a - The US invasion of Iraq b - IPv6 c - DNSSec d - All of the above


IPv6 is Incompatiblewith IPv4 on the Wire!

The Insanity and Short-Sighted Arrogance of this is Utterly Mind Blowing


Let’s Dispel Some Myths


Myth: IPv4 is Running Out• IPv4 Free Pool run-out in a few years• This is in line with the graphs of Frank

Solensky over ten years ago• IPv4 will go to a Trading Model• Registries will become Title Agents,

not allocators, of IPv4 space• RIRs developing full multi-RIR/LIR

open source software to certify and verify title to IPv4 and IPv6 resources


Myth: IPv6 Transition is Easy• IPv6 was designed with no serious

thought to operational transition• IPv6 is on-the-wire incompatible

with IPv4• Might have been avoided, e.g. if IPv6

had variable length addressing, IPv4 might have become the 32 bit variant

• There are no simple, useful, scalable translation or transition mechanisms


Myth: IPv6 Eliminates NATs• An IPv6-only site can not reach IPv4

Internet because it can not source packets from an IPv4 address

• There will be significant IPv4-only Internet for a decade or more

• All IPv6 sites will need IPv4 space and will have NATs with ALGs

• IPv6 increases NAT use in short and medium term, i.e. a decade or more


Myth: IPv6 Reduces Routing Load

• Multi-homing in IPv6 is the same as in IPv4, there is no new routing model

• Traffic engineering in IPv6 is the same as in IPv4, no new TE model

• Enterprises will slice and dice their IPv6 /32s to handle branches etc.

• The routing table will fragment more and more over time


Myth: IPv6 Space is Infinite• 64 bits goes to every LAN• This leaves half the bits gone!• Some folk use /64 for Point-to-Point!• RIRs are giving away /32s• In 15 years we will think of these as

we now think of legacy /8s in IPv4 space

• We once thought 32 bits was enough


Myth: IPv6 is more Secure• IPv6 does nothing IPv4 does not,

though it promised to• IPSec is the recipe in either case• IPSec does not work well in a mixed

IPv4/IPv6 environment (think VPN from an IPv4-only hotel room)

• It is true that address space scanning will be somewhat harder

• Ha Ha, think botnet scanning and a black market in hot space


Myth: Incremental Deployment

• For an enterprise, the entire chain, from database back end, through applications, through firewalls, to the border router must all support v6 or the enterprise can not deploy

• For ISP, provisioning systems, monitoring, measurement, billing, …

• And everyone needs support from all their vendors


Myth: Routers Fully Support IPv6

• But not 100% in hardware• Especially not if you add ACLs• Folk do not know this because there is

no good IPv6 traffic test equipment• And all vendors are not spinning the

ASICs to solve this• Not all v4 features are supported over

IPv6: MIBs, SNMP over v6, …


Myth: No Static Numbering

• IPv6 Auto Configuration is not widely used in enterprise as security policy prefers known (i.e. DHCP) addresses

• Similarly, ISP backbone addresses and customer addresses must be known for logging, audit, CALEA, …


Myth: IPv6 is Deployed• Pioneers are still moving cautiously• Early adopters are just starting to

enter the game• Actual measured traffic is very

small (so it makes routers look as if they can handle the traffic)

• But there are success stories


Prefix Allocation Distribution


BGP Prefix Announcements


Myth: IPv6 Ubiquitous in Japan• NTT.com and IIJ do have dual-stack on

leased line services. And NTT.com has dual-stack on ADSL.

• The main high-bandwidth layer-2 service is NTT-East/West B-Flets, for ETTH

• Neither NTT nor IIJ provide native or dual-stack over B-Flets

• NTT-East/West do have a VoIP and TV-to-STB closed IPv6 universe over B-Flets. No Internet.


Myth: IPv6 Will Replace IPv4• Not given current lack of universal

vendor support from back end to border router

• It is far easier to use NAT and IPv4• IPv4 with NATs requires no new

expense, conversion, training, …• This is architecturally horrible, it is

just financial reality


The Reality• “96 more bits, no magic”

-- Gaurab Raj Upadhaya• But we definitely need more bits!• The key questions are how to use

them?• How to transition without losing

anyone or anything?


What Can We Do?

IPv6Deployment

IPv4Free Pool

$/IPv4/24

Today

MakeMakeThisThisEasyEasy


How?• Identify current transition problems• See that they are fixed• Ask the IETF to fix the outstanding

protocol issues• Ask IETF to stop embellishing so that

we can deploy something stable• Push vendors to support IPv6 and the

tools for us to transition


What We Should NotNot Do• Pretend that there are no transition

problems. It just makes things harder.

• Give away IPv6 space in strange ways to “promote” IPv6. IPv4 run-out will promote IPv6 for us.

• Make messes we will have to live with forever.


Areas of Concern/Study• Global Issues• Administrative Infrastructure• Layers 1 and 2• Backbone Engineering• Last Mile/Kilometer• Consumer/SOHO Self-Installed CPE• Enterprise• Server Farm• Campus• Exchange Points• Applications• Telephony• More?


Layers 1 and 2• DOCSIS 3.0 for Cable

– CMTS support lacking– Massive installed base of DOCSIS 2

modems• 802

– All media protocols support IPv6– While the protocols support IPv6, this

does not at all mean that implementations do


Backbone Engineering• Core Routing – conversion to dual

stack is slow• Provisioning, Address Assignment,

DNS, ...• DHCPv6 and DNS Integration• Monitoring and Measurement over

v6?• New line cards are often required!


Last Kilometer• Authentication and session setup• Conversion to IPoE, DHCP expensive• Provisioning, back-end database, ...• “How to scale the routing/provisioning

combo to deal with million of customers using stable prefix delegation?”


Consumer Self-Installed CPE

• $50 DSL Modems do not support v6• $50 Firewalls do not support v6• Teredo does not really scale• shim6 is does not solve enterprise or

large site, and is not deployable due to security and routing model issues


Security Devices

• Dave Piscitello made a presentation “IPv6 Support Among Commercial Firewalls” at the last ARIN meeting<http://www.arin.net/meetings/minutes/ARIN_XX/PDF/thursday/Firewalls_Piscitello.pdf>

• Less than 1/3 had IPv6 Transport• 25% supported IPv6 Routing• And it gets Worse from there


Enterprise

• Databases, PeopleSoft, Siebold, Business Applications, ...

• Firewalls, VPNs, Access, ...• Millions of lines of in-house code• NFS Appliances, unknown• Load Balancers


Applications• Where is the web page with matrix of

application by platform showing which are v6 capable and clickable link on how to turn it on?

• http://www.deepspace6.net/docs/ipv6_status_page_apps.html out of date

• Many applications which support v6 have sufficiently poor performance that early adopters are being told to turn v6 off


SMTP: An Example• Email/SMTP is a mandatory application• Everyone needs to be able to send email to

arbitrary recipients, i.e. everyone else• But, due to SPAM, no one can run an open

SMTP relay• So all IPv6 sites need to have the ability to

SMTP to arbitrary IPv4 sites• Therefore everyone needs private dual

stack relay until the world is all dual stack SMTP

[ example by Jeffrey Streifling ]


Why is Japan in Better Shape?

• Folk with vision (i.e. Murai) convinced the government that early movement to IPv6 was wise for Japan

• Government $upport$ IPv6 research• Government $upport$ IPv6

development by industry, vendors, …• Government give$ tax incentive$ to

enterprises which become v6 compatible


What Can We Do?


Summary• No More Bull❨^&*• No More Excuses• Shut up and Spend the Money

-- Lucy, in a stressed moment


Principle: One Internet

• Under no circumstances can we allow the Internet to fragment

• During transition, everybody still needs to talk to everyone else at will

• And it would be good if the End to End principle could be kept as much as possible


Principle: Dual Stack

• The core needs to be dual IPv4/IPv6 during all of transition or kludges will escalate horrifyingly

• The further dual stack goes toward the edge (enterprise, net services, consumer, …) the easier it will be

• Configuration, Management, and Measurement need to be simplified


Five Phases• Denial, from both ‘sides’:

– We can ignore brain-dead IPv6– IPv6 is perfect and those greedy

fools just have to deploy it• Dual stack with IPv4 Dominant• Dual stack with both widely used• Dual stack with IPv6 Dominant• The IPv6 Internet (getting ready

for IPv10 transition:)


Pressure on Routing• IPv4 address space price escalation

and the consequential NATs will put serious new pressure on routing

• If it takes a $10m router to deal with 2m routes and churn then 96% of ISPs die and enterprises can not be DFZ multi-homed

• So all sized routers, from enterprise border to ISP core, need to handle >2m routes with churn


Do Not Hack!• Do not accept hacks around the

routing scaling problem such as tunneling from enterprise border to some $10m ‘core’ router

• Think TLA/NLA and be fearful • Think ten Monopoly ISPs and be

very very fearful


Forwarding is Behind• Because of lack of market, it will be five

years before all major router vendors support dual stack at line rate with ACLs

• Some vendors are not even spinning the ASICs for all platforms and line cards

• Needs to be all vendors because ISPs can not be vendor-locked by transition

• So we are not interested in “We can do it, they can’t” marketing wars.


Good IPv6 Test Equipment

• Router/Switch vendors claim wonderful performance

• But you can not test it because there is a serious lack of good test/exercise equipment


Stop Adding Features• Stop trying to market IPv6 through

more and more kinky features• IPv4 free pool run-out will either sell

IPv6 or there will be an IPv4 NAT world

• Adding features just gives vendors and operators reasons to delay

• Freeze the damned thing and give us a chance to deploy it!


ULA: A Bad Example• Because ULA is address-based• ‘Borders’ need to filter packets• To not leak and not accept leaks, needs

both source and destination filters• Do not make special address space

– Remember 240/4 and that clean-up– IPv6 space is supposed to be infinite!

• Give them real IPv6 space and tell them to just not announce it to the DFZ


Principle: NATs

• End to End Principle is very desirable• But IPv6 on the wire is incompatible

with IPv4• During transition there will be NATs• Get over it• But we need to make it so they can

fade away and not be there forever


NAT-PT• At the edge, the enterprise, consumer,

etc. need to run IPv6 but need to talk to both IPv4 and IPv6 services

• When IPv6 becomes dominant, the IPv4 sites will still need to talk to the then predominantly IPv6 Internet

• The IETF needs to standardize 4/6 NAT for ICMP, UDP, TCP, DNS, SMTP, HTTP, SIP, RTP, and maybe an API of how ALGs plug in


IETF and Reality• In July 2007, the IETF published RFC

4966 “Reasons to Move the Network Address Translator – Protocol Translator (NAT-PT) to Historic Status”

• This tells you a lot about the IETF, their level of operational clue, and how much they care about religion as opposed to IPv6 deployabilty


Windows XP• XP can move payload over IPv6• But does not do DNS over IPv6

transport (bad bug! “Buy Vista”)• So, the LAN has 1918 IPv4 space

to carry DNS, but no exit for IPv4• You get an IPv6 and IPv4 Address

but should use the IPv6 for all real transport as IPv4 has no default


NAT-PT & Security• DNSsec has to terminate on the

NAT if translating and use ALG• IPSec can transit NAT-PT• DNS, SMTP, HTTP, SIP, RTP ALGs

will be critical• IPsec must be made easy for users

to configure


The DNS Hack• On a pure IPv6 network, if I get an

A record, what do I do?• Panic, you can’t use an A record• So the local DNS Cache has a hack,

totd, which takes an A, embeds it within a hacked IPv6 prefix, and synthesizes an AAAA

• NAT-PT knows the hack prefix, and strips it back to IPv4 to dual-stack


Picture by Vasaka Visoottiviseth - mahidol.ac.thtotd – DNS Trans Proxy


NANOG Experiment

• 2008.02.17-20 in San Jose, CA

• 550 attendees, all on laptops

• 45mb Exit to Net

• Wireless with multiple SSIDs

• Dual Stack, V6-only, Hacked V6 for XP


Local-4Global-6

No DefaultIPv4 Route

NAT-PT

RA

totdDNS

XP

IPv6-Only Backbone

IPv6-onlyRouter

UNIX Mac

Dual-Stack

Internet

IPv6 Users

Multiple Pink ClientNetworks are Hungfrom an IPv6-OnlyBackbone Network

Legacy Users

NANOG/APRICOT/…IPv6 Experiment

Crutches

Vista Linux


NANOG Experiment• All SSIDs were up from start

– nanog and nanog-a fully dual-stack– nanog-v6 with NAT-PT & tots– nanog-v6-xp with 1918

• Tuesday, for an hour or so, nanog and nanog-a went away

• We gathered stats from NAT-PT, net use, AP binding, traffic, …

• Reported on Wednesday


What We Learned• ¾ of users said they could get on net but

it was actually more like half

• Every component except UNIXes had bugs: NAT-PT, Vista, MacOS, ...

• The prize to MacOS which dropped capital A from DNS server entry

• We hope vendors now working to fix



Ongoing Experiments• Similar, but not identical, experiments

[will be] held at– APRICOT/APNIC last week

– IETF next week, different goals

– AfNOG, LACNIC, ...

– ICANN, ...


Summary• IETF

– NAT-PT– No More ‘Features’ or hacks, e.g. ULA

• Vendors– Dual Stack on the Fast Path with ACLs– 2+m Routes with churn on all routers– Test Equipment

• ISPs– Dual Stack to the Customer Edge

• Governments: incent, don’t regulate


You Can Help!


How You Can Help

http://www.civil-tongue.net/6and4/

write to [email protected] you can contribute

Please!


Thanks To

ISOC

Internet Initiative Japan

Lucy Lynch, Rob Austein,Russ Housley, Steve Bellovin &

Jari Arko

Documents

IPv6 Transition & Operational Realitycenic08.cenic.org/program/slides/080312.cenic-v6-op-reality.pdfIPv6 Transition & Operational Reality CENIC 08 Oakland / 2008.03.12 Randy Bush